LLM Course

LLM Course: setup notes

This kit holds setup notes only. Mindket wrote them. The course itself, its images and its links stay in Maxime Labonne's repository: https://github.com/mlabonne/llm-course

These notes are not made or endorsed by him. The README calls the project "LLM Course" and the repository's description reads "Course to get into Large Language Models (LLMs) with roadmaps and Colab notebooks." The notes were written from the README and the repository page, read on 3 November 2026.

License: Apache-2.0, as the repository's license field and its LICENSE file (https://github.com/mlabonne/llm-course/blob/main/LICENSE) say. The LICENSE file is the standard Apache 2.0 text. Its appendix still reads "Copyright [yyyy] [name of copyright owner]", so it names no holder. The pages and videos by other people that the README links to belong to their own owners, and the README's disclaimer says "I am not affiliated with any sources listed here."

What the course is

The repository holds three entries: README.md, a LICENSE file and an img folder (the repository page, 3 November 2026). The README is the course. It says the course is divided into three parts:

1. LLM Fundamentals is optional and covers mathematics, Python and neural networks.
2. The LLM Scientist focuses on building the best possible LLMs using the latest techniques.
3. The LLM Engineer focuses on creating LLM-based applications and deploying them.

Each part has a roadmap image. Each section inside a part has a short paragraph, its topics in bold and a list of references (articles, videos, papers, repositories) with a one-line note on each. A section near the top, "Notebooks", lists notebooks and articles the author wrote, with Open In Colab buttons. The README says the LLM course will always stay free, and that the author co-wrote the LLM Engineer's Handbook, a book based on the course. Source: https://github.com/mlabonne/llm-course/blob/main/README.md

The sections

Read from the README on 3 November 2026. The Fundamentals sections sit inside a toggle that GitHub shows closed, and the README says you might not want to start there but refer to it as needed.

LLM Fundamentals (optional)

| # | Section | Topics the README names |
|---|---|---|
| 1 | Mathematics for Machine Learning | Linear algebra, calculus, probability and statistics |
| 2 | Python for Machine Learning | Python basics, data science libraries, data preprocessing, machine learning libraries |
| 3 | Neural Networks | Fundamentals, training and optimization, overfitting, implementing a multilayer perceptron |
| 4 | Natural Language Processing (NLP) | Text preprocessing, feature extraction techniques, word embeddings, recurrent neural networks |

The LLM Scientist

| # | Section | Topics the README names |
|---|---|---|
| 1 | The LLM Architecture | Architectural overview, tokenization, attention mechanisms, sampling techniques |
| 2 | Pre-Training Models | Data preparation, distributed training, training optimization, monitoring |
| 3 | Post-Training Datasets | Storage and chat templates, synthetic data generation, data enhancement, quality filtering |
| 4 | Supervised Fine-Tuning | Training techniques, training parameters, distributed training, monitoring |
| 5 | Preference Alignment | Rejection sampling, direct preference optimization, reward model, reinforcement learning |
| 6 | Evaluation | Automated benchmarks, human evaluation, model-based evaluation, feedback signal |
| 7 | Quantization | Base techniques, GGUF and llama.cpp, GPTQ and AWQ, SmoothQuant and ZeroQuant |
| 8 | New Trends | Model merging, multimodal models, interpretability, test-time compute |

The LLM Engineer

| # | Section | Topics the README names |
|---|---|---|
| 1 | Running LLMs | LLM APIs, open-source LLMs, prompt engineering, structuring outputs |
| 2 | Building a Vector Storage | Ingesting documents, splitting documents, embedding models, vector databases |
| 3 | Retrieval Augmented Generation | Orchestrators, retrievers, memory, evaluation |
| 4 | Advanced RAG | Query construction, tools, post-processing, program LLMs |
| 5 | Agents | Agent fundamentals, agent protocols, vendor frameworks, other frameworks |
| 6 | Inference optimization | Flash Attention, key-value cache, speculative decoding |
| 7 | Deploying LLMs | Local deployment, demo deployment, server deployment, edge deployment |
| 8 | Securing LLMs | Prompt hacking, backdoors, defensive measures |

Ways to read it

• setup/github.md: read the README on GitHub, or clone the repository.
• setup/google-colab.md: the fine-tuning, quantization and other notebooks, each with its Colab link and, where the README gives one, its article.
• setup/tool-notebooks.md: the eight tool notebooks, two of which name RunPod and a free ZeroGPU.
• setup/more-from-the-author.md: the DeepWiki page, the book, the blog and the other places the README points to.

What you need first

• To read: a web browser. The README is a page on GitHub.
• To clone: Git. The clone address is in setup/github.md.
• For the notebooks: the Open In Colab buttons open them in Google Colab. The README gives no steps beyond the links, and Mindket did not open the notebooks.
• Nothing in the repository installs or runs on your computer. The README has no install or run commands.

Good to know

• The README gives each section a paragraph and a list of references. Most of what you learn comes from the pages, videos and notebooks it links, each under its owner's terms.
• The README gives no prices for the notebooks. One notebook row says it works in a free-tier Google Colab, and another says it uses RunPod.
• Two rows, LazyMergekit and Create MoEs with MergeKit, carry the same Colab link in the README.
• Mindket's request to the DeepWiki page the README points to was stopped by a security check page (HTTP 429) on 3 November 2026, so it was not read.
• To remove it, delete the clone folder. Nothing else is installed.

Sources

• README: https://github.com/mlabonne/llm-course/blob/main/README.md
• LICENSE: https://github.com/mlabonne/llm-course/blob/main/LICENSE
• The repository page and its GitHub API answer (license, clone address): https://github.com/mlabonne/llm-course and https://api.github.com/repos/mlabonne/llm-course